This talk evaluates how generative AI, multimodal models, and agentic AI are moving from research prototypes into clinical research and trial operations. Key topics include drug development, site selection, RWE/RWD, patient identification, evidence generation, and workflow automation. The talk discusses practical use cases, frameworks, and tools for AI validation, explainability, uncertainty, and trustworthy deployment. The session closes with regulatory expectations and responsible adoption of generative AI in clinical development.

While the industry rapidly adopts AI to optimize trial design and execution, the greatest hurdle is no longer technological—it is behavioral. Drawing on over 27 years of experience navigating landscape shifts, this session explores the critical intersection of AI implementation, change management, and behavioral science. I will share strategic insights from the past 2 years leading AI transformation. It will detail how to move beyond pilot programs by building an internal "Community of Practice", upskilling clinical teams, and utilizing behavioral science principles to overcome resistance.

This conversation will explore how agentic AI can transform end-to-end clinical trial design, from protocol development and endpoint selection to country and site strategy, patient recruitment planning, and operational readiness. It examines where specialized AI agents can augment cross-functional teams, automate complex work processes, and enable faster, more adaptive decision-making—ultimately with a goal of accelerating overall drug development timelines while preserving scientific rigor, regulatory alignment, data quality, and human oversight.

AI implementation in clinical research must move beyond experimentation to sustained, measurable impact. This presentation offers a practical journey from identifying high-value ideas and aligning scientific and business priorities to funding, building, and deploying fit-for-purpose AI tools in real workflows. It addresses validation, governance, change management, and value measurement, while emphasizing the critical success factor—adoption. Attendees will gain a pragmatic framework for creating trusted AI solutions that are used, scaled, and sustained.

Designing and running a clinical trial today means coordinating hundreds of interdependent decisions such as eligibility criteria, site selection, visit schedules, data flows, patient funnel, external feedback, etc. We'll share how our organization is using AI to model these decisions together, generating optimized protocols and orchestrating downstream execution in near real time. This session covers the architecture behind the approach, lessons from early deployments, and what it means for cross-functional collaboration.

Clinical stakeholders traditionally relied on static reports and dashboards, creating technical bottlenecks. Now, Large Language Models (LLMs) enable users to query clinical databases directly using natural language. This AbbVie case study explores the implementation of conversational capabilities to democratize data access. We demonstrate how generative AI empowers non-technical users to generate immediate insights, accelerate decision-making, and bridge the gap between complex clinical data and actionable operational intelligence.

Regeneron will share how SOP Assistant streamlines GxP Procedural Document change management through an innovative, end-to-end automated solution powered by Agentic AI. This streamlined automation-supported business process reduces manual effort by approximately 50%, accelerates Global Development decision-making, and enables faster identification of future agency change impacts. Cross-functional collaboration, strong governance, product-focused delivery, and continuous improvement have shaped the vision of a true SOP Assistant.

Determining which documents a clinical trial needs for ethics and regulatory submissions by country is slow, manual, and error-prone. The Keystone Project predicts a list of documents by pairing curated regulatory intelligence with AI-driven confidence scoring using probabilistic and rules-based machine learning. This application gives study teams instant, auditable answers across 80+ countries and 125 documents. This talk covers how Keystone was built, validated, and its impact in study start-up.

A natural CLARA sequel from 2026 covering additional capabilities and user experience. Features to highlight will be MSTeams integration, industry benchmarking, new datasets, enrollment prediction, alerts, and next-best-action recommendations. We will also highlight the governance model implemented to ensure trust & accuracy. We would also cover some themes of “Humans in the loop with AI” which will speak to how we’re managing downstream user experience with our tool.

Clinical trial planning generates thousands of decisions across emails, meetings, presentations, and trackers, yet much of the rationale behind those decisions is difficult to reconstruct later. This presentation will showcase an AI-powered approach developed through a collaboration between Boehringer Ingelheim and Microsoft to automatically capture project change history ("churn"), create structured decision summaries, and establish a foundation for future predictive analytics. Attendees will see how AI can transform fragmented project information into organizational knowledge.

The framework combines automated evaluation metrics with risk-stratified Subject Matter Expert (SME) assessment to provide a comprehensive and scientifically rigorous approach for evaluating AI-generated content. The framework acknowledges the limitations of deterministic metrics and emphasizes alignment between quantitative measures and expert judgment. A key component of the framework is the rigorous design of human-in-the-loop evaluation studies. We describe approaches for sample size determination, non-inferiority testing of AI-generated versus human-generated responses, inter-rater reliability assessment, and quantification of agreement between automated metrics and expert evaluation. Bayesian ordinal mixed-effects models are explored as a methodology for powering and analyzing comparative evaluation studies. By establishing consistent, evidence-based validation practices, this framework enables the responsible deployment of GenAI solutions within a regulated environment. It supports faster and more efficient evaluation workflows, provides greater confidence in AI-assisted decision-making, and helps Pfizer scale trustworthy AI capabilities that accelerate innovation and drug development.

AI is rapidly moving from experimentation to deployment across clinical research, yet many organizations still struggle to measure value in a credible and repeatable way. This presentation describes a practical framework for evaluating AI initiatives by linking operational improvements to business outcomes through acceleration, productivity, cost savings, and cost avoidance—covering establishing baselines, avoiding common ROI pitfalls, and creating scalable approaches for prioritizing and governing AI investments.

Making the best use of AI in clinical analytics requires more than setting up an MCP server. A strong governance strategy; a well-designed semantic layer; a flexible, dynamic user interface; a thoughtful approach to integration into user workflows; and a pre-established evaluation plan are all essential for any project's success. Providing the strategy and tools to show end users "what's in it for them" leads to faster uptake and impact.

Clinical operations teams rely on validated dashboards, but many study-specific questions still require manual joins, spreadsheet work, and interpretation across different levels of data. This presentation shares how an agentic AI analyst was designed to answer natural-language clinical trial operations questions. The session will demonstrate how AI can turn operational metrics into faster, traceable decision support.

SAP2TLFis a human-in-the-loop agentic AI system that accelerates clinical trial statistical programming for Tables, Listings, and Figures. It generates study-aware SAS analytics code grounded in sources such as SAPs, TLF shells, SDTM/ADaM metadata, and curated examples, with syntax and executability checks before human review. Built-in provenance, audit logging, versioning, and end-to-end traceability enable governed, evidence-based code generation that shortens timelines, reduces rework, and supports responsible, auditable AI adoption.